{ "id": 14397, "title": "Motivation of Bot Operators", "url": "/competitions/facebook-recruiting-iv-human-or-bot/discussion/14397", "author_name": "", "post_date": "2015-05-27T06:52:04.220Z", "votes": 1, "comment_count": 7, "views": 1304, "content": "

Has anyone else considered that there might be two distinct motivations at play?

\n

1) Bot owner wants to win the auction

\n

2) Bot owner wants to increase to winning price of he auction while not actually winning it (e.g. is trying to profit from selling an item)

\n

Each motivation would drive different behaviors, and while I've never heard of a penny auction site allowing users to sell items (like ebay),  we dont know for sure that its not the case.   

\n

Motivation 2 would explain certain behavior, such as a bidder bidding over themselves.

", "messages": [ { "id": "80101", "postDate": "05/27/2015 06:52:04", "content": "

Has anyone else considered that there might be two distinct motivations at play?

\n

1) Bot owner wants to win the auction

\n

2) Bot owner wants to increase to winning price of he auction while not actually winning it (e.g. is trying to profit from selling an item)

\n

Each motivation would drive different behaviors, and while I've never heard of a penny auction site allowing users to sell items (like ebay),  we dont know for sure that its not the case.   

\n

Motivation 2 would explain certain behavior, such as a bidder bidding over themselves.

", "rawMarkdown": "", "votes": null }, { "id": "80257", "postDate": "05/28/2015 15:33:42", "content": "

My knee-jerk response was not being sure why it would matter. But thinking about it more, it does lead to a few ideas on feature selection such as opening bid, closing bid, consecutive bids, etc. At first glance, consecutive bids appears to be more weighted towards humans actually, but I only ran a quick script to check (so entirely likely/feasible it has some error in it leading to that conclusion). Regardless, would make for potential features to try (whether they are any good, haven't tested yet).

", "rawMarkdown": "", "votes": null }, { "id": "80289", "postDate": "05/28/2015 18:03:57", "content": "

I also ran a script to figure out how many bidders over bid themselves only. The results were overwhelmingly human. I wonder if there is some kind of threshold that the bids have to beat before an auction is finalised?

", "rawMarkdown": "", "votes": null }, { "id": "80305", "postDate": "05/28/2015 21:00:19", "content": "

It may also be the case that the auction closes at certain time unless somebody makes a bit, in that case we move the end of the auction by few minutes. You may also look at the number of bets from the end of the auction or from the beginning. 

", "rawMarkdown": "", "votes": null }, { "id": "80366", "postDate": "05/29/2015 15:05:42", "content": "

First pass through testing these and being the opening bid or the closing bid, or having consecutive bids, don't seem to contribute enough to matter in anything I test. (at least vs other features in my test).

\n

(I am only testing in R's random forest, using a random selection of 300k bids, and varImpPlot to see - so take that for what it is worth)

", "rawMarkdown": "", "votes": null }, { "id": "80385", "postDate": "05/29/2015 16:25:53", "content": "

Are you running that through the bid data or some aggregates on a bidder level?

", "rawMarkdown": "", "votes": null }, { "id": "80393", "postDate": "05/29/2015 17:10:00", "content": "

Bidder level of whether or not they have consecutively bid in the same auction (and a few similar things like that), and then at the bid level of whether that bid was the first or last in its auction. Could certainly try more aggregates/variations, but in this particular context, they were totally bottoming out in the varImpPlot thing vs more obvious things like totalIPs, totalBids, etc.

\n

Again, take all of that with a huge grain of salt - more just observations of what I have seen with my code, vs anything definite - definitely don't assume I have any idea of what I am doing.

", "rawMarkdown": "", "votes": null }, { "id": "80408", "postDate": "05/29/2015 18:58:18", "content": "

As a general rule, dummy variables do not provide enough information, you need to look into some other statistics. Also often one statistic does not provide much information about the distribution of a feature.

", "rawMarkdown": "", "votes": null } ], "comments": [ { "id": 80257, "author_name": "omgponies", "author_url": "", "post_date": "05/28/2015 15:33:42", "content": "

My knee-jerk response was not being sure why it would matter. But thinking about it more, it does lead to a few ideas on feature selection such as opening bid, closing bid, consecutive bids, etc. At first glance, consecutive bids appears to be more weighted towards humans actually, but I only ran a quick script to check (so entirely likely/feasible it has some error in it leading to that conclusion). Regardless, would make for potential features to try (whether they are any good, haven't tested yet).

", "votes": null, "replies": [] }, { "id": 80289, "author_name": "abhishekbarnwal", "author_url": "", "post_date": "05/28/2015 18:03:57", "content": "

I also ran a script to figure out how many bidders over bid themselves only. The results were overwhelmingly human. I wonder if there is some kind of threshold that the bids have to beat before an auction is finalised?

", "votes": null, "replies": [] }, { "id": 80305, "author_name": "mpekalski", "author_url": "", "post_date": "05/28/2015 21:00:19", "content": "

It may also be the case that the auction closes at certain time unless somebody makes a bit, in that case we move the end of the auction by few minutes. You may also look at the number of bets from the end of the auction or from the beginning. 

", "votes": null, "replies": [] }, { "id": 80366, "author_name": "omgponies", "author_url": "", "post_date": "05/29/2015 15:05:42", "content": "

First pass through testing these and being the opening bid or the closing bid, or having consecutive bids, don't seem to contribute enough to matter in anything I test. (at least vs other features in my test).

\n

(I am only testing in R's random forest, using a random selection of 300k bids, and varImpPlot to see - so take that for what it is worth)

", "votes": null, "replies": [] }, { "id": 80385, "author_name": "mpekalski", "author_url": "", "post_date": "05/29/2015 16:25:53", "content": "

Are you running that through the bid data or some aggregates on a bidder level?

", "votes": null, "replies": [] }, { "id": 80393, "author_name": "omgponies", "author_url": "", "post_date": "05/29/2015 17:10:00", "content": "

Bidder level of whether or not they have consecutively bid in the same auction (and a few similar things like that), and then at the bid level of whether that bid was the first or last in its auction. Could certainly try more aggregates/variations, but in this particular context, they were totally bottoming out in the varImpPlot thing vs more obvious things like totalIPs, totalBids, etc.

\n

Again, take all of that with a huge grain of salt - more just observations of what I have seen with my code, vs anything definite - definitely don't assume I have any idea of what I am doing.

", "votes": null, "replies": [] }, { "id": 80408, "author_name": "mpekalski", "author_url": "", "post_date": "05/29/2015 18:58:18", "content": "

As a general rule, dummy variables do not provide enough information, you need to look into some other statistics. Also often one statistic does not provide much information about the distribution of a feature.

", "votes": null, "replies": [] } ], "raw_markdown_by_id": { "80101": "", "80257": "", "80289": "", "80305": "", "80366": "", "80385": "", "80393": "", "80408": "" }, "source": "meta" }